Date: Wed, 5 Nov 2003 10:55:46 -0600
Reply-To: Thompson Bill T Contr USAFSAM/FEC <Bill.Thompson@BROOKS.AF.MIL>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Thompson Bill T Contr USAFSAM/FEC <Bill.Thompson@BROOKS.AF.MIL>
Subject: Re: Transforming data
Content-Type: text/plain; charset="iso-8859-1"
Peter,
You are correct in your assumptions and I will follow your lead regarding
the residuals, skew and kurtosis. If the residuals are "normally"
distributed would you suggest NOT transforming the data?
One issue that has come up regarding "eye" is that these are "fellow" eyes
from the same person (humans). Since the results are "highly" correlated at
baseline it was suggested we just pool the eyes and eliminate that variable.
Unfortunately, subjects received surgery on one eye at a time with the
possibility of several weeks between surgeries. As a result the followup
data is not as highly correlated because of different surgical outcomes for
different eyes. Hence, we are trying to decide if we should analyze each
eye separately or use a mixed model.
Your thoughts.
Bill
-----Original Message-----
From: Peter Flom [mailto:flom@NDRI.ORG]
Sent: Tuesday, November 04, 2003 2:21 PM
To: SAS-L@LISTSERV.UGA.EDU
Subject: Re: Transforming data
Bill
If I understand yoiu correctly, you have measurements of one variable
(light sensitivity) at each of 36 combinations of 4 other variables.
Light senstivity is your DV, and time, light, screen and eye are your
IVs (Please correct me if I am mistaken here).
In this case, you certainly cannot transform the DV in some cells and
not others. But the key question is not whether the values in each of
the 36 combinations is normally distributed, the question is whether the
residuals from your model are normally distributed.
I have found that the best way of looking at this is graphcially, but I
don't have SAS/Graph, and I usually do this in R. If you have SAS
Graph, there is doubtless a good way to make some charts.
Alternatively, you can do tests and look at the skew and kurtosis of
the residuals.
BTW, if 'eye' is left vs. right on the same person (or whatever sort of
creature you are examining) then I would recommend you look at a mixed
model, since this will be nested within person
HTH
Peter
Peter L. Flom, PhD
Assistant Director, Statistics and Data Analysis Core
Center for Drug Use and HIV Research
National Development and Research Institutes
71 W. 23rd St
www.peterflom.com
New York, NY 10010
(212) 845-4485 (voice)
(917) 438-0894 (fax)
>>> Thompson Bill T Contr USAFSAM/FEC <Bill.Thompson@brooks.af.mil>
11/4/2003 3:09:59 PM >>>
Peter,
These 36 variables are actually part of a 3x3x2x2 repeated measures
design.
time x light x screen x eye
These variables represent "contrast sensitivity" values taken at
baseline,
and then 12 months and 24 months post refractive surgery. The issue
has
been raised as to wether or not we should analyze the "contrast
sensitivity"
values or the "log contrast sensitivity" values. These are continuous
data.
The data was analyzed as a repeated measures design described above
using
the "raw" contrast sensitivity data.
